Imagine a valley so wide the other rim looks like a watercolor wash, pale and hazy, and in the middle of the valley lay a floor where sunlight feels thin. Most ambitious projects trace that shape. New software snarls service tickets before the patches land, lean initiatives jam factory lines before processing time improves, and every bright-eyed AI pilot will mis-label invoices before it automates anything. The dip feels irrational until you recall that the valley is not a bug in progress. It is the toll we pay for change.
My professor recently forwarded a note to the cohort: “New research from MIT Sloan documents the ‘WBB’ — Worse-Before-Better — dynamic in AI adoption by firms … and, as our framework predicts, the depth of the ‘worse’ period is larger for older, less digitally mature organizations.” The article traced tens of thousands of U.S. manufacturing plants from 2017 to 2021, finding that productivity slipped 1.33 percentage points on average after an AI rollout, and ballooned to nearly 60 points once the authors corrected for optimistic self-selection. The valley, it warned, yawns widest where routines have calcified. I skimmed the email twice, felt the curve settle behind my eyes, then let the message sit in my inbox like an uncashed check.
Later that same day a close confidant, the former head of R & D for pharmaceutical interventions whose therapies have saved more lives than I will ever meet, forwarded Ethan Mollick’s Substack dispatch, “The Bitter Lesson versus The Garbage Can.” Mollick paints organizations as “chaotic garbage cans” where problems and solutions collide at random and recounts a CEO who, after seeing a true process map, put his head on the table and groaned, “This is even more fucked up than I imagined.” Sutton’s Bitter Lesson — AI self-learning outpacing hard-coded human expertise — hovers over the anecdote, and my confidant’s note in his email seals the point: “My bet is on the messy model.” Different inboxes, same crooked “U.” Time to cash that check.
If artificial intelligence will stumble, plain old operations has been paying the fare for years. Since turnaround specialist Brian Niccol slipped into Starbucks’ corner office last September, the coffee giant has been tearing up its playbook: paring menus, redesigning stores, and unleashing “Green Apron Service,” a hospitality drill that vows 80 percent of drinks in under four minutes that will march through 11,000 U.S. shops by mid-August. Two-day rallies for 14,000 managers, thicker staffing rosters, and wage bumps north of half a billion dollars are part of the tab. So nobody attuned to the WBB curve should flinch at this quarter’s headlines: U.S. traffic down for a sixth straight quarter and net income plunging 47 percent to $558 million. The J-curve is simply doing its job.
The landmarks of that valley rarely change. First comes the honeymoon, a rollout party with glossy forecasts. Next arrives the ugly middle, when legacy and pilot run in parallel: two billing systems, two recipe queues, two mental models tugging at the same bandwidth. Costs double, throughput stumbles, and morale frays. Only after yesterday’s workflow shuts down while new muscle memory takes hold does performance climb toward the far rim.
Why do managers keep falling to greater canyon depths than could otherwise be expected? Three culprits surface again and again.
Learning-curve overhead swallows hours that never appear on a blueprint.
Resource cannibalization forces teams to bankroll both the old and the new until the switchover is safe.
Cultural drag keeps bright people welded to familiar tools long after the upgrade is live.
Scholars have also mapped ways out of the canyon. Nelson Repenning and John Sterman (two leading researchers in this area I’m lucky enough to count as my professors) show that leaders under quarterly pressure often cut the very training spend that would shorten the slog. They recommend instrumenting the valley with leading indicators such as queue length or cycle-time variance so hidden gains surface before profit does. Erik Brynjolfsson, Daniel Rock, and Chad Syverson describe the Productivity J-Curve, where intangible capital soaks up resources first and shows up in the metrics only later; their cure is to ring-fence capacity so the right-hand ascent is funded rather than starved. Finally, practiced operators storyboard the climb, showing investors the whole “U,” time-stamping milestones, and repeating that the dip is tuition, not failure.
The valley is no shallower for being charted. But a good map steadies the nerves. Wherever you stand on that downward tilt, may the next step tilt a shade higher, and may you keep the budget, the patience, and the curiosity to keep walking.